JP2022074801A - Dose distribution creation program taking influence of magnetic field into account, dose distribution creation method taking influence of magnetic field into account, and dose distribution creation device - Google Patents

Abstract

To provide a dose distribution creation program taking an influence of a magnetic field into account, a dose distribution creation method taking an influence of a magnetic field into account, and a dose distribution creation device for executing dose calculation under a magnetic field faster and more accurately than ever before.SOLUTION: A dose distribution creation program taking an influence of a magnetic field into account causes a computer to acquire a captured image taken of a subject with a magnetic field and a radio wave or a photon beam, dose distribution under a non-magnetic field, and dose distribution under a magnetic field, and to input the dose distribution under a non-magnetic field and information based on the captured image to a learning model consisting of a plurality of layers that has performed prior learning using the dose distribution under a non-magnetic field, dose distribution under a magnetic field, and the information based on the captured image, and to estimate the dose distribution under a magnetic field.SELECTED DRAWING: Figure 7

Description

The present invention relates to a dose distribution creating program considering the influence of a magnetic field, a method for creating a dose distribution considering the influence of a magnetic field, and a dose distribution creating device.

As a new treatment device for radiotherapy, MR-Linac, which integrates an MR (Magnetic Resonance) device using a magnetic field and a radiotherapy device (Linac), has begun to spread (see, for example, Non-Patent Document 1).

In such radiotherapy using MR-Linac, for example, first, before treatment, an image is taken with an MRI (Magnetic Resonance Imaging) device, and a treatment plan is created based on the taken MR image. Next, an MR image is acquired on the day of treatment with an MR-Linac device. Next, using the MR image taken on the day of treatment, the irradiation position and the like in the planned treatment plan are calculated, and the doctors confirm whether or not the treatment plan needs to be revised. Then, if the treatment can be carried out according to the planned treatment plan, the radiation therapy according to the treatment plan is carried out. If the treatment cannot be performed according to the planned treatment plan, the treatment plan is created and cured while the subject (patient) is on the treatment table using the MR image taken on the day of the treatment.

As described in Non-Patent Document 1, in a device in which an MR device using such a magnetic field and a radiotherapy device are integrated, there is an influence of the magnetic field on radiation. For example, secondary electrons emitted from the phantom are bent by a magnetic field and re-enter the phantom, causing an Electron Return Effect (ERE) phenomenon in which the dose at the interface between air and water increases. In addition, since the secondary electrons are bent by the magnetic field, the phenomenon that the dose distribution is biased including the high dose region occurs. Therefore, in the treatment using such a device, it is necessary to calculate the behavior of the radiation beam in the patient body in a state where a magnetic field is constantly generated.

In addition, a system for formulating a radiotherapy treatment plan through the use of a machine learning approach and neural network components in the preparation of a radiotherapy plan has been proposed (see, for example, Patent Document 1). In the technique described in Patent Document 1, a neural network is trained using one or more 3D medical images, one or more 3D anatomical maps, and one or more dose distributions. , Predict fluence maps or dose maps. In the technique described in Patent Document 1, during training, the neural network receives an expected dose distribution determined by the neural network to be compared with the expected dose distribution. In the technique described in Patent Document 1, the comparison is repeated until a predetermined threshold value is achieved. The technique described in Patent Document 1 then utilizes a trained neural network to provide a three-dimensional dose distribution.

Special Table 2019-526380 Gazette

Tomoyuki Kakutani, "Current Status of MR-Linac Fusion Device", <Special Feature of the 110th Annual Meeting of the Japanese Society of Medical Physics> Symposium 2 "Fusion Technology of Diagnosis and Treatment", Medical Physics Vol. 36, No. 4, 2016, p229-235

However, most conventional treatment devices do not support magnetic fields. There are some treatment planning devices that support magnetic fields, but it was necessary to calculate each radiation in order to reflect the bending force of the magnetic field in the dose distribution. For this reason, in the prior art, it is necessary to use a slow calculation algorithm for accurate dose calculation, it takes a lot of time to calculate the dose, and there is a problem that the subject stays on the treatment table for a long time. ..

The present invention has been made in view of the above problems, and is a dose distribution creation program considering the magnetic field effect, which enables dose calculation under a magnetic field faster and more accurately than before, and a dose distribution considering the magnetic field effect. It is an object of the present invention to provide a preparation method and a dose distribution preparation device.

In order to achieve the above object, the dose distribution creation program considering the influence of the magnetic field according to one aspect of the present invention includes a photographed image in which a subject is photographed by a magnetic field and radio waves or photon rays, and a dose distribution under a non-magnetic field. A learning model consisting of a plurality of layers learned by acquiring the dose distribution under a magnetic field, the dose distribution under a non-magnetic field, the dose distribution under a magnetic field, and information based on the captured image. Is input with the dose distribution under the non-magnetic field and the information based on the captured image, and the dose distribution under the magnetic field is estimated.

Further, in the dose distribution creation program considering the influence of the magnetic field according to one aspect of the present invention, the information based on the captured image is the electron density generated from the CT image of the subject, the MR image of the subject, and the CT image of the subject. It may be one of a map and an electron density map generated from the MR image of the subject.

Further, in the dose distribution creation program considering the influence of the magnetic field according to one aspect of the present invention, the dose distribution consists of a distribution parallel to the irradiation direction of radiation or a plurality of tomographic images perpendicular to the irradiation direction, and is pre-learned and The dose distribution under the non-magnetic field used for input is the normalized distribution or multiple tomographic images, and the dose distribution under the magnetic field is normalized and then each pixel value is corrected by a coefficient representing the influence of the magnetic field. It may be a plurality of tomographic images obtained.

Further, in the dose distribution creation program considering the influence of the magnetic field according to one aspect of the present invention, the learning model is a convolutional neural network provided with a Dense Block, and the computer is a Dense of the learning model composed of the plurality of layers. Any layer after Block may allow the subject to input information based on an image taken by a magnetic field and radio waves or photon rays.

Further, in the dose distribution creation program in consideration of the influence of the magnetic field according to one aspect of the present invention, the learning model may be a convolutional neural network including Skip Connection.

In order to achieve the above object, in the method for creating a dose distribution in consideration of the influence of a magnetic field according to one aspect of the present invention, a computer takes a photographed image of a subject by a magnetic field and radio waves or photon rays, and a dose distribution under a non-magnetic field. A learning model consisting of a plurality of layers learned by acquiring the dose distribution under a magnetic field, the dose distribution under a non-magnetic field, the dose distribution under a magnetic field, and information based on the captured image. The dose distribution under the non-magnetic field and the information based on the captured image are input to, and the dose distribution under the magnetic field is estimated.

In order to achieve the above object, the dose distribution creating device according to one aspect of the present invention includes a photographed image in which a subject is photographed by a magnetic field and radio waves or photon rays, a dose distribution under a non-magnetic field, and a dose distribution under a magnetic field. The non-magnetic field is applied to a learning model consisting of a plurality of layers learned using the acquisition unit, the dose distribution under the non-magnetic field, the dose distribution under the magnetic field, and the information based on the captured image. It is provided with a magnetic field correction unit that inputs the lower dose distribution and information based on the captured image and estimates the dose distribution under a magnetic field.

According to the present invention, it is possible to calculate the dose under a magnetic field that is faster and more accurate than before.

It is a figure which shows the outline of the dose distribution creation processing which considered the influence of the magnetic field which concerns on embodiment. It is a block diagram which shows the structural example of the dose distribution creation apparatus which concerns on embodiment. It is a figure for demonstrating the CT value electron density conversion table. It is a figure which shows an example of the magnetic field correction model which concerns on embodiment. It is a figure which shows the processing example at the time of learning which concerns on embodiment. It is a flowchart which shows the processing procedure example of the learning of the learning model which concerns on embodiment. It is a flowchart which shows the example of the estimation processing procedure of the dose distribution under the magnetic field which concerns on embodiment. It is a figure which shows the learning curve. It is a figure which shows the dose error. It is a figure which shows the 1st example of a dose distribution. It is a figure which shows the example of the profile corresponding to FIG. It is a figure which shows the 2nd example of a dose distribution. It is a figure which shows the example of the profile corresponding to FIG. It is a figure which shows the Gamma passing rate at all the measurement points. It is a figure which shows the Gamma passing rate in the skin (3 mm). It is a figure which shows the example of another learning model which concerns on embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings used in the following description, the scale of each member is appropriately changed in order to make each member recognizable. In the following description, a case where the apparatus includes an MR photographing apparatus, a CT photographing apparatus, and an MR-Linac will be described as an example, but the method of the present embodiment can be applied to other devices and systems.

<Outline of dose distribution creation process considering the influence of magnetic field>
FIG. 1 is a diagram showing an outline of a dose distribution creation process in consideration of the influence of a magnetic field according to the present embodiment. As shown in FIG. 1, the dose distribution creation process considering the influence of the magnetic field includes a learning step and a dose distribution creation step.

In the learning step, the MR imaging device or the CT imaging device provided in the dose distribution creating device captures an image (photographed image) g11 (MR image or CT image of the subject) under non-magnetization or a magnetic field. Next, the dose distribution creating device calculates and acquires the captured image g11 and the dose distribution g12 under a non-magnetic field and the dose distribution g13 under a magnetic field from the captured image. Next, the dose distribution creating device separates the dose distribution g12 under a magnetic field and the dose distribution g13 under a non-magnetic field into three-dimensional to two-dimensional information (g15). Next, the dose distribution creating device obtains the two-dimensional non-magnetic field dose distribution, the two-dimensional magnetic field dose distribution, and the electron density map (ED map) information g14 or the captured image g11 generated from the captured image g11. Input to the learning model g16 to perform supervised learning. In the following description, the "electron density map (ED map) information generated from the captured image or the captured image" is also referred to as "information based on the captured image". The dose distribution consists of a distribution parallel to the irradiation direction of radiation (a parallel three-dimensional distribution, a plurality of vertical two-dimensional distributions, etc.) or a plurality of vertical tomographic (two-dimensional) images. Further, the dose distribution may be a dose distribution calculated by using a photon beam, a dose calculated by using a charged particle beam, or another dose distribution. It is also a non-magnetic field dose distribution, normalized distribution or multiple tomographic images input during pre-learning and treatment, and the dose distribution under magnetic field is normalized and then each pixel value is affected by the magnetic field. It is a plurality of tomographic images obtained by correcting with a representative coefficient (for example, the maximum pixel value under a magnetic field / the maximum pixel value under a non-magnetic field). The dose distribution creating device may input the dose distribution g12 under a non-magnetic field and the dose distribution g13 under a magnetic field into the learning model g16 as they are in three dimensions without dividing them into three-dimensional to two-dimensional information. .. Further, the dose distribution creating device tunes the learning model g16 so as to match the dose distribution under the magnetic field which is the teacher data. As described above, in the present embodiment, the learning model is trained using the information based on the captured image, the dose distribution under the non-magnetic field, and the dose distribution under the magnetic field.

In the dose distribution creating step, the MR-Linac included in the dose distribution creating device acquires a photographed image under non-magnetization or a magnetic field on the day of treatment. The dose distribution creation device calculates and acquires the dose distribution under a non-magnetic field from the captured image. Next, the dose distribution creating device inputs the dose distribution g17 under a non-magnetic field divided into two dimensions and the information based on the captured image into the trained learning model g16. Next, the dose distribution creating device creates a dose distribution g18 under a magnetic field by the learned learning model g16.

<Configuration example of dose distribution creation device>
Next, a configuration example of the dose distribution creation device will be described. FIG. 2 is a block diagram showing a configuration example of the dose distribution creating device according to the present embodiment. As shown in FIG. 2, the dose distribution creating device 1 includes a gantry device 11, a bed device 12, an imaging device 13, a radiotherapy device 14, an image processing unit 15 (acquisition unit, processing unit), and a dose distribution processing unit 16 (acquisition unit). , Processing unit), magnetic field correction unit 17, control unit 18, storage unit 19, operation unit 20, and display unit 21. Further, the magnetic field correction unit 17 includes a learning model 171.

The dose distribution creating device 1 includes, for example, an imaging device such as an MR imaging device, a CT imaging device, and an MR-Linac.

The gantry device 11 is a device that photographs a subject. An imaging device 13 and a radiotherapy device 14 are attached to the gantry device 11.

The sleeper device 12 is, for example, a device on which the subject lies. The sleeper device 12 may be a bed or the like that is manually moved by the photographer (at least one of the photographer and the imaging assistant), and the control unit 18 may control the operation.

The photographing device 13 is an MR photographing device or a CT (Computed Tomography) photographing device. The imaging device 13 photographs the patient under the control of the control unit 18, and outputs the captured image to the image processing unit 15. The captured image is an MR image or a CT image under non-magnetization or a magnetic field.

The radiation therapy device 14 is, for example, a linear accelerator (Linac). The radiotherapy device 14 performs treatment based on the treatment plan according to the control of the control unit 18.

The image processing unit 15 acquires a photographed image taken by the photographing apparatus 13. The image processing unit 15 generates ED map information from the acquired captured image, normalizes the generated ED map information, and inputs it to the learning model 171 of the magnetic field correction unit 17. Alternatively, the image processing unit 15 normalizes the acquired captured image and inputs it to the learning model 171 of the magnetic field correction unit 17.

The dose distribution processing unit 16 acquires a captured image from the imaging device 13. The dose distribution processing unit 16 performs a dose distribution calculation on the acquired captured image and acquires a dose distribution under a non-magnetic field and a dose distribution under a magnetic field for each gate. At the time of learning, the dose distribution processing unit 16 uses the dose distribution under a non-magnetic field and the dose distribution under a magnetic field. Further, at the time of treatment, the dose distribution processing unit 16 may only calculate the dose distribution under a non-magnetic field and may not calculate the dose distribution under a magnetic field. The gates are, for example, 1 gate, 2 opposite gates, 2 orthogonal gates, 2 obliquely incident gates, 4 gates, cross fire, multiple gates, etc. in the fixed irradiation method. Further, the dose distribution processing unit 16 calculates the dose distribution under a non-magnetic field by, for example, an AAA (Anaisotropic Anisotropy Algorithm) method, a CCC (Collapsed Convolution) method, a Monte Carlo method, or the like. Further, the dose distribution processing unit 16 calculates the dose distribution under a magnetic field by, for example, the Monte Carlo method. The dose distribution processing unit 16 performs the following first processing and second processing for three-dimensional information, and further divides the information into two dimensions.
First process; normalize the dose distribution in a non-magnetic field to a value between 0 and 1, for example.
Second process; the dose distribution under magnetic field is normalized to a value between 0 and 1, for example, and multiplied by a coefficient {for example (maximum value (under magnetic field) / maximum value (under non-magnetic field))}.
The dose distribution processing unit 16 does not have to divide the three-dimensional information into two-dimensional information.
The dose distribution processing unit 16 sets the lower limit of the table to 0 and sets the upper limit to 0 in the range of the CT-ED Table based on the CT-ED Table (CT-relative electron density conversion table) used for the dose calculation stored in the storage unit 19. Normalize as 1 to create an ED map. The ED map is two-dimensional information.

At the time of learning, the magnetic field correction unit 17 uses the dose distribution under the magnetic field as teacher data, and trains the learning model 171 using the information based on the captured image and the dose distribution under the non-magnetic field. The training method will be described later. At the time of treatment, the magnetic field correction unit 17 inputs the dose distribution under the normalized non-magnetic field and the ED map information or the captured image generated from the normalized captured image, and uses the trained learning model for each gate. Correct the dose distribution. A configuration example of the learning model 171 will be described later.

The control unit 18 controls each functional unit and each device by using the operation information acquired by the operation unit 20 and the information stored in the storage unit 19.

The storage unit 19 stores a CT-ED Table (CT value electron density conversion table). The shooting conditions and the like included in the operation result acquired by the operation unit 20 are stored. The storage unit 19 stores a control program, a threshold value, and the like.

The operation unit 20 is, for example, a touch panel sensor, a mechanical switch, a keyboard, a mouse, etc. provided on the display unit 21. The operation unit 20 detects the operation result operated by the user and outputs the detected operation result to the control unit 18.

The display unit 21 is, for example, a liquid crystal image display device, an organic EL (Electroluminescence) image display device, or a portable terminal such as a tablet. The display unit 21 displays the presentation image output by the control unit 18.

The image processing unit 15, the dose distribution processing unit 16, the magnetic field correction unit 17, and the control unit 18 may be, for example, a personal computer, a CPU (central processing unit), a DSP (digital signal processor), or the like.

<Electron density map>
Here, the electron density map (ED map) will be described. FIG. 3 is a diagram for explaining a CT value electron density conversion table.
The value that directly affects the dose calculation is the electron density. The CT value electron density conversion table is a map (g23) that converts the CT value (MR image g21) or MR value (MR image g22) of each voxel into electron density. In the map g23, the horizontal axis is the CT value [HU] and the vertical axis is the electron density [g / cm ³ ]. The image processing unit 15 uses this conversion table to convert a CT image into an electron density to create an electron density map.

<Magnetic field correction model>
Here, an example of a magnetic field correction model, which is a learning model, will be described. FIG. 4 is a diagram showing an example of a magnetic field correction model according to the present embodiment. The magnetic field correction model 100, which is the learning model of FIG. 4, is an example based on CNN (Convolutional Neural Networks) and DenseNet (Reference 1). Further, in the magnetic field correction model 100 of the present embodiment, since the emphasis is placed on local features, DownSampling such as pooling and compression is not used.

Reference 1: Gao Huang, Zhuang Liu et al., “Densely Connected Convolutional Networks”, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2017

The magnetic field correction model 100 is connected, for example, in the order of convolution layer 121, Dense Block 122, Dense Block 123, Dense Block 124, convolution layer 125, convolution layer 126, convolution layer 127, and convolution layer 128.

A two-dimensional non-magnetic field dose distribution 101 and a two-dimensional ED map 102 are input to the magnetic field correction model 100. The resolution of the ED map 102 is, for example, 160 × 224.
Further, the magnetic field correction model 100 re-inputs the ED map 102 at an arbitrary position after the Dense Block, for example, between the convolution layer 125 and the convolution layer 126.

First, a convolution with a filter size of 3 × 3 is performed on the input (105).
Next, the Dense Blocks 122 to 124 are processed by the Dense Blocks (106 to 108).
Next, convolution with a filter size of 1 × 1 is performed (109).
Next, with the output of the convolution layer 125 and the ED map 102 as inputs, convolution with a filter size of 3 × 3 is performed (110).
Next, convolution with a filter size of 3 × 3 is performed (111 to 112).
Next, convolution with a filter size of 1 × 1 is performed (113).
The output of the magnetic field correction model 100 is a dose distribution 114 under a two-dimensional magnetic field, and the resolution is, for example, 160 × 224.

The diagram of the region indicated by reference numeral 131 is an image diagram of Dense Block. In the Dense Block, all the outputs from each front layer are used as inputs to the rear layers. In addition, Dense Block combines layers of the same feature size in order to maximize the transmission of information between layers, and inputs the output of the layer before a certain layer in order to maintain back propagation. Further, the Dense Block alternately includes 1 × 1 convolution and 3 × 3 convolution called BottleNeck.

In the example shown in FIG. 4, an example in which the ED map is input to an arbitrary layer after the Dense Block is described, but the input to the arbitrary layer after the Dense Block is a CT image (or MR image) of the subject or It may be an image obtained by normalizing a CT image (or MR image) and cutting it out in two dimensions.

The configuration example of the magnetic field correction model 100 and the Dense Block shown in FIG. 4 is an example, and the number of layers, the number of filters, and the like are not limited to this. Further, in the example shown in FIG. 4, the example in which the network is CNN has been described, but the network may be another network, for example, an RNN (Recurrent Neural Network) or the like.

The magnetic field correction model g100 of the present embodiment is provided with a Dense Block, and the ED map 102 is input again at an arbitrary position after the Dense Block. In Dense Block, for example, the influence of secondary electrons under a magnetic field is corrected. Thereby, according to the present embodiment, the anatomical structure can be reflected.

<Processing during learning>
Next, a processing example at the time of learning will be described. FIG. 5 is a diagram showing a processing example at the time of learning according to the present embodiment. In the following processing example, an electron density map (ED map) information is used as information based on the captured image, and an example in which two-dimensional information cut out from three dimensions is input to the learning model will be described.

(1) Shooting process g301
The photographing device 13 captures a captured image g302 (MR image or CT image) under a non-magnetic field or a magnetic field. Next, the dose distribution creating device 1 acquires a captured image. Next, the dose distribution processing unit 16 calculates and acquires the dose distribution under a non-magnetic field and the dose distribution under a magnetic field from the acquired captured image.

(2) Pretreatment step g311
The dose distribution creating device 1 uses a learning model (magnetic field correction) of the separated two-dimensional dose distribution g312 under a non-magnetic field and the electron density map (ED map) information g313 generated from the captured image using the CT value electron density conversion table. Model) Input to g322. Further, the dose distribution g314 under a magnetic field is used as the teacher data.

(3) Learning and confirmation process g321
The dose distribution creating device 1 performs learning and confirmation by a 4-partition cross-validation method using a 4-partitioned learning model g323.

(4) Test process g331
The dose distribution creating device 1 synthesizes and averages the estimated dose distribution g332 under the magnetic field for each of the four divided learning models g323 to create the corrected dose distribution g333 under the magnetic field.
The dose distribution creating device 1 compares with the dose distribution g314 under a magnetic field, which is the correct answer data, creates a reward based on the comparison result, and reflects it in learning. Since the dose distribution g314 under a magnetic field, which is the correct answer data, is learning, there is no problem even if it takes time to calculate. Therefore, the behavior of the radiation beam in the patient is calculated in a state where the magnetic field is always generated.

In the process of FIG. 5, the above-mentioned process such as normalization is omitted.

<Method>
Next, an example of the work procedure will be described. First, the processing procedure at the time of learning will be described.
FIG. 7 is a flowchart showing an example of a procedure for estimating the dose distribution under a magnetic field according to the present embodiment. In the following processing example, the electron density map (ED map) information is used as the information based on the captured image, and the two-dimensional information cut out from the three dimensions is input to the learning model.

(Step S11) The operation unit 20 acquires shooting conditions and the like for learning.
(Step S12) The photographing device 13 photographs a subject under the control of the control unit 18 for learning.
(Step S13) The image processing unit 15 acquires a captured image (CT layer or MR image) captured by the imaging device 13. Subsequently, the image processing unit 15 generates an electron density map from the captured image using the CT value electron density conversion table, and normalizes the generated electron density map.

(Step S14) The dose distribution processing unit 16 acquires a captured image from the imaging device 13. Subsequently, the dose distribution processing unit 16 calculates and acquires the dose distribution under a non-magnetic field and the dose distribution under a magnetic field from the captured image. Subsequently, the dose distribution processing unit 16 normalizes the dose distribution under a magnetic field and the dose distribution under a non-magnetic field in three dimensions, and further divides the dose distribution into two dimensions.

(Step S15) The magnetic field correction unit 17 uses an electron density map generated from an image captured by the image processing unit 15 and a non-magnetic field dose distribution and a magnetic field dose distribution calculated by the dose distribution processing unit 16. Then, the learning model 171 is trained.

Next, an example of the procedure at the time of treatment will be described.
FIG. 6 is a flowchart showing an example of a processing procedure according to the present embodiment. In the following processing example, an electron density map (ED map) information is used as information based on the captured image, and an example in which two-dimensional information cut out from three dimensions is input to the learning model will be described.

(Step S21) The operation unit 20 acquires imaging conditions and the like for treatment. The imaging device 13 photographs the subject under the control of the control unit 18 for treatment. Note that this process may be omitted. The dose distribution creating device 1 may, for example, acquire and use an image taken in advance.

(Step S22) The image processing unit 15 acquires a photographed image (CT layer or MR image) photographed by the photographing apparatus 13 or a photographed image previously photographed. Subsequently, the image processing unit 15 generates an electron density map from the captured image using the CT value electron density conversion table, and normalizes the generated electron density map.

(Step S23) The dose distribution processing unit 16 acquires a captured image from the imaging device 13. Subsequently, the dose distribution processing unit 16 calculates and acquires the dose distribution under a non-magnetic field from the captured image. Subsequently, the dose distribution processing unit 16 normalizes the dose distribution under a non-magnetic field in three dimensions, and further divides the dose distribution in two dimensions.

(Step S24) The magnetic field correction unit 17 inputs the dose distribution under the non-magnetic field and the electron density map generated from the captured image into the learning model 171 to estimate the dose distribution under the magnetic field.

(Step S25) The control unit 18 causes the display unit 21 to display the estimated dose distribution under the magnetic field.

In the above-mentioned processing, the dose distribution processing unit 16 does not have to convert the dose distribution from three-dimensional to two-dimensional during learning or treatment. In this case, the dose distribution processing unit 16 may normalize the three-dimensional dose distribution and input it into the learning model 171.
Further, in step S24 of the above process, the control unit 18 may create a treatment plan based on the estimated dose distribution under the magnetic field, and display the created treatment plan on the display unit 21.

<Confirmation result>
Next, the confirmation results of the learning of the learning model of the present embodiment and the case of estimation using the learned learning model will be described.
FIG. 8 is a diagram showing a learning curve. In FIG. 8, the horizontal axis is the number of learnings (number of epochs), and the vertical axis is MSE (Mean Squared Error) [× 10 ^-3 ]. The polygonal line g401 is the loss characteristic of the training data, and the polygonal line g402 is the loss characteristic of the verification data. As shown in FIG. 8, the MSE has converged after several learnings.

FIG. 9 is a diagram showing a dose error. In FIG. 9, the vertical axis is MSE) [× 10 ^-3 ]. Reference numeral g411 is a dose error when corrected by the present embodiment using Dense Block for CNN, and reference numeral g412 is a dose error when no correction is made. As shown in FIG. 9, the MSE is improved to 1/3 or less when corrected by the method of the present embodiment.

Next, an example of the first dose distribution and profile will be described.
FIG. 10 is a diagram showing a first example of a dose distribution. The dose distribution g421 shows the dose distribution under a non-magnetic field without correction, and the enlarged view g422 is an enlargement of a part of the dose distribution g421. The dose distribution g431 shows the dose distribution under a magnetic field without correction, and the enlarged view g432 is an enlargement of a part of the dose distribution g431. The reference numeral g441 indicates ED map. The dose distribution g451 shows the dose distribution under a magnetic field corrected by the present embodiment, and the enlarged view g452 is an enlargement of a part of the dose distribution g451. As shown in FIG. 10, when the correction according to the present embodiment is performed, the distribution of the lower right portion is different from that without the correction as shown in the enlarged view g452.

Next, the profile corresponding to FIG. 10 will be described.
FIG. 11 is a diagram showing an example of a profile corresponding to FIG. 10. In FIG. 10, the horizontal axis is a pixel, the left vertical axis of the graph g460 is a relative dose, the right vertical axis of the graph g460 is an error, and the vertical axis of the graph g470 is a relative electron density.
Further, the solid line g461 shows the relative dose in the dose distribution under the magnetic field, the broken line g462 shows the relative dose in the dose distribution under the magnetic field corrected by the method of the present embodiment, and the broken line g463 shows the dose distribution under the uncorrected magnetic field. The relative dose in is shown. Further, the dashed line g464 indicates an error in the magnetic field dose distribution corrected by the method of the present embodiment, and the dashed line g465 indicates an error in the uncorrected magnetic field dose distribution.

It should be noted that the pixel of about 90 or more corresponds to the lower right of the enlarged portion of the dose distribution in FIG.
As shown in FIG. 11, when the correction is performed by the method of the present embodiment, there is a difference between the relative dose and the error as compared with the case where the pixel is corrected at about 90 or more. The case of correction by the method of the present embodiment is much closer to the relative dose in the dose distribution under the magnetic field of the solid line g461 even if the number of pixels is about 90 or more as compared with the case of no correction. Also, regarding the error, if the pixel is 90 or more and the error is not corrected, the error is increasing.

Next, an example of the second dose distribution and profile will be described. The second dose distribution is an example of an image of the intestine.
FIG. 12 is a diagram showing a second example of the dose distribution. The dose distribution g521 shows the dose distribution under a non-magnetic field without correction, and the enlarged view g522 is an enlargement of a part of the dose distribution g521. The dose distribution g531 shows the dose distribution under a magnetic field without correction, and the enlarged view g532 is an enlargement of a part of the dose distribution g531. The reference numeral g541 indicates ED map. The dose distribution g551 shows the dose distribution under a magnetic field corrected by the present embodiment, and the enlarged view g552 is an enlargement of a part of the dose distribution g551. As shown in FIG. 12, when the correction according to the present embodiment is performed, the distribution of the lower portion is different from that without the correction as shown in the enlarged view g552.

Next, the profile corresponding to FIG. 12 will be described.
FIG. 13 is a diagram showing an example of a profile corresponding to FIG. 12. In FIG. 13, the horizontal axis is the pixel number (position), the left vertical axis of the graph g560 is the relative dose, the right vertical axis of the graph g560 is an error, and the vertical axis of the graph g570 is the relative electron density.
Further, the solid line g561 shows the relative dose in the dose distribution under the magnetic field, the broken line g562 shows the relative dose in the dose distribution under the magnetic field corrected by the method of the present embodiment, and the broken line g563 shows the dose distribution under the uncorrected magnetic field. The relative dose in is shown. Further, the dashed line g564 indicates an error in the magnetic field dose distribution corrected by the method of the present embodiment, and the dashed line g565 indicates an error in the uncorrected magnetic field dose distribution.

It should be noted that about 60 to 70 pixels are the boundary line portion of the intestine, and correspond to the portion below the enlarged portion of the dose distribution in FIG.
As shown in FIG. 13, when the method of the present embodiment is used for correction, there is a difference between the relative dose and the error as compared with the case where the pixels are not corrected at about 60 to 70. The case of correction by the method of the present embodiment is closer to the relative dose in the dose distribution under the magnetic field of the solid line g561 even if the number of pixels is about 60 to 70 as compared with the case of no correction. Also, regarding the error, the number of errors increases more when the pixels are 60 to 70 and the error is not corrected.

Next, an example of the gamma analysis result will be described with reference to FIGS. 14 and 15. In addition, Gamma passing rate means that the two dose distributions completely match if it is 100%.
FIG. 14 is a diagram showing a Gamma passing rate at all measurement points. FIG. 15 is a diagram showing a Gamma passing rate on the skin (3 mm). The horizontal axis of FIGS. 14 and 15 is (ΔD [%] / Δd [mm]), and the vertical axis is Gamma passing rate [%]. In addition, ΔD [%] is a dose difference, and Δd [mm] is DTA (Disstance To Agreement). Further, the hatched area g601 is a case where the correction of the present embodiment is performed, and the hatching area g602 is a case where the correction is not performed.

As shown in FIGS. 14 and 15, it can be seen that the correction is performed by the method of the present embodiment and is in good agreement (close to 100%) with the dose distribution with a magnetic field, which is the true value.

Conventional general calculation algorithms cannot take into account the effects of magnetic fields. When the Monte Carlo algorithm was used, the influence of the magnetic field could be taken into consideration, but the calculation time was about 10 to 20 minutes. For this reason, in the conventional technique, it is difficult to make a treatment plan by waiting for the result of such arithmetic processing while lying on the bed for the subject at the time of treatment. And, a highly accurate and high-speed dose calculation algorithm under a magnetic field does not currently exist.
On the other hand, according to the present embodiment, it is possible to perform correction in consideration of the influence of the magnetic field at a higher speed than before.

<Examples of other learning models>
Next, an example of another learning model will be described.
FIG. 16 is a diagram showing an example of another learning model according to the present embodiment. As shown in FIG. 16, the magnetic field correction model 100A, which is a learning model, is a CNN provided with a Skip Connection (identity map). The Skip Connection has a bond that jumps over the layers. In addition, each layer of the magnetic field correction model 100A is a 3 × 3 convolution layer, a 1 × 1 convolution layer, a 3 × 3 pooling (pruning, Pool; Pooling) layer, and a deconvolution (Deconv; Deconvolution) layer. ) Layer.

Also when this learning model is used, as in the above-described embodiment, the teacher data is the dose distribution under a magnetic field during training, and the input is the dose distribution g701 under a non-magnetic field and the electron density map generated from the captured image. It is g702 (or a photographed image). The inputs at the time of treatment are the dose distribution g701 under a non-magnetic field and the electron density map g702 (or the captured image) generated from the captured image. The output of the learning model is the dose distribution g703 under a magnetic field, similar to the embodiment described above.

Also when this magnetic field correction model 100A is used, learning is performed in advance in the same manner as in the above-described embodiment, and the dose distribution g703 under the magnetic field is estimated. Even when this magnetic field correction model 100A is used, the dose distribution g703 under the magnetic field can be estimated faster and more accurately than before.
The configuration shown in FIG. 15 is an example, and the number of layers, input / output resolution, and the like are not limited to this.

In the above-mentioned method, the dose distribution is a charged particle beam for both the dose distribution when taking a CT image by X-rays (photon rays) and the dose distribution when taking an MR image by a magnetic field and radio waves. It is applicable to both the dose distribution calculated using the dose distribution and other dose distributions.
The method of this embodiment can also be used to derive the final dose distribution used for treatment and can be coupled to an optimization process.

A program for realizing all or part of the functions of the dose distribution creating device 1 in the present invention is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into the computer system. By performing this, all or part of the processing performed by the dose distribution creating device 1 may be performed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. Further, the "computer system" shall also include a WWW system provided with a homepage providing environment (or display environment). Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM, and a storage device such as a hard disk built in a computer system. Furthermore, a "computer-readable recording medium" is a volatile memory (RAM) inside a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, it shall include those that hold the program for a certain period of time.

Further, the program may be transmitted from a computer system in which this program is stored in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the "transmission medium" for transmitting a program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. Further, the above program may be for realizing a part of the above-mentioned functions. Further, a so-called difference file (difference program) may be used, which can realize the above-mentioned function in combination with a program already recorded in the computer system.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

1 ... Dose distribution creation device, 11 ... Stand device, 12 ... Sleep device, 13 ... Imaging device, 14 ... Radiation therapy device, 15 ... Image processing unit, 16 ... Dose distribution processing unit, 17 ... Magnetic field correction unit, 18 ... Control Unit, 19 ... Storage unit, 20 ... Operation unit, 21 ... Display unit, 171 ... Learning model

Claims (7)

On the computer
The subject was made to acquire a photographed image taken by a magnetic field and radio waves or photon rays, a dose distribution under a non-magnetic field, and a dose distribution under a magnetic field.
A learning model consisting of a plurality of layers learned using the dose distribution under the non-magnetic field, the dose distribution under the magnetic field, and the information based on the captured image, the dose distribution under the non-magnetic field, and the above. Input information based on the captured image and estimate the dose distribution under a magnetic field.
A dose distribution creation program that takes into account the effects of magnetic fields. The information based on the captured image is
It is one of a CT image of the subject, an MR image of the subject, an electron density map generated from the CT image of the subject, and an electron density map generated from the MR image of the subject.
The dose distribution creation program in consideration of the influence of the magnetic field according to claim 1. The dose distribution is
It consists of a distribution image parallel to the irradiation direction of radiation or a plurality of tomographic images perpendicular to the irradiation direction, and the dose distribution under the non-magnetic field used for pre-learning and input is the normalized distribution image or a plurality of tomographic images. Yes, the dose distribution under a magnetic field is a plurality of tomographic images obtained by correcting each pixel value with a coefficient representing the effect of the magnetic field after normalization.
The dose distribution creation program considering the influence of the magnetic field according to claim 1 or 2. The learning model is
A convolutional neural network with Dense Block,
The computer
Any layer after the Dense Block of the learning model consisting of the plurality of layers is made to input information based on an image taken by a photon beam or a charged particle beam to the subject.
The dose distribution creation program in consideration of the influence of the magnetic field according to any one of claims 1 to 3. The learning model is
A convolutional neural network with Skip Connection,
The dose distribution creation program in consideration of the influence of the magnetic field according to any one of claims 1 to 3. The computer
The photographed image of the subject taken by a magnetic field and radio waves or photon rays, the dose distribution under a non-magnetic field, and the dose distribution under a magnetic field are acquired.
A learning model consisting of a plurality of layers learned using the dose distribution under the non-magnetic field, the dose distribution under the magnetic field, and the information based on the captured image, the dose distribution under the non-magnetic field, and the above. Input information based on the captured image and estimate the dose distribution under a magnetic field.
A method for creating a dose distribution that takes into account the effects of magnetic fields. An acquisition unit that acquires a photographed image of a subject taken by a magnetic field and radio waves or photon rays, a dose distribution under a non-magnetic field, and a dose distribution under a magnetic field.
The dose distribution under the non-magnetic field, the dose distribution under the magnetic field, and the information based on the captured image are used in a learning model composed of a plurality of layers learned using the dose distribution under the non-magnetic field, and the dose distribution under the non-magnetic field. A magnetic field correction unit that estimates the dose distribution under a magnetic field by inputting information based on the captured image,
A dose distribution creation device equipped with.

Images